SLOPE: A MATLAB Revival Amber L. Stuver stuver@gravity.psu.edu

Outline Introduction How Does it Work? Results Future Work Description Trigger Generation & Clustering Results Data Conditioning False rates Efficiency and false alarms Future Work 19 August 2004 Penn State University

Introduction SLOPE: Motivation: is a simple algorithm is not very computationally expensive was previously applied to S1 analysis issues with background rate excluded it from upper limit estimation time domain algorithm Motivation: - Burst S1 paper 19 August 2004 Penn State University

Description SLOPE is a linear filter that returns the best fit slope to data The data is windowed according to the expected timescale of a burst signal This sliding window is moved through the data An event will typically cause a cascade of triggers These triggers are clustered together to form the candidate event 19 August 2004 Penn State University

Trigger Generation SLOPE previously did not apply an adaptive threshold This fixed threshold was attributed to the background estimation problems in S1 SLOPE now applies a threshold based on the idealized white noise probability that a slope that passes threshold is not accidental This probability threshold relates directly to the expected false rate for white noise 19 August 2004 Penn State University

Trigger Clustering If two triggers are separated by less than the width of the window, the two triggers are considered part of the same event These clusters are returned as candidate events The data used in this illustration is Gaussian noise with zero mean and unit variance. 19 August 2004 Penn State University

Data Conditioning Since SLOPE is a time domain ETG, data conditioning is necessary uses the same data conditioning as BlockNormal The data is down sampled and Kalman filtered to remove narrow line features base banded regressed to remove power lines and calibration lines whitened Results presented here are in the band between 512 to 640 Hz This band is the cleanest before data conditioning. 19 August 2004 Penn State University

False Rates S2 playground false rates track the ideal white noise rate Departures from the ideal false rate indicate non-stationarities False rates can be improved with applying SLOPE to stationary epochs 19 August 2004 Penn State University

Efficiencies GravEn was used to simulate sine-Gaussians at 153 Hz, Q=10 Simulations were injected at an average rate of 1 Hz Criteria for detection: trigger cluster must overlap with the time of maximum amplitude of simulated signal Efficiency: False Alarm: 19 August 2004 Penn State University

Efficiency vs. False Alarm The average peek amplitude of simulation was on the order of 10-20. 19 August 2004 Penn State University

Future Work SLOPE is to be included in my thesis investigation on the efficacies of the different burst ETG’s for different waveform properties Break data up into stationary epochs to further improve false rate and efficiency Investigate: effects of window size on efficiency different clustering criteria (i.e. different minimum distance between triggers) Suggestions welcome! 19 August 2004 Penn State University